Speaker
Description
JWST and HST in space and sensitive (sub)millimetre interferometers on the ground have transformed our view of distant galaxies from faint fuzzy blobs to rich resolved information across the electromagnetic spectrum, tracing young and old stars, ionised gas and dust. This introduces the exciting prospect of finally uncovering where and how the build-up of stars proceeded “within” galaxies, and how the relative distribution of stars and obscuring dust clouds changed over time. Dust is an important ingredient to the star formation cycle, but its attenuating effects on light have also long plagued efforts to reconstruct galaxies' growth histories. Long-wavelength sensitivity promises to overcome these hurdles, but with better data comes a need for better modelling tools too. 3D radiative transfer represents a powerful technique capable of translating in a self-consistent, physically motivated manner the characteristics of an input model galaxy to what an observer would see with the aforementioned telescopes. Ideally, one would like to turn this process around and constrain what family of model galaxies (i.e., what amounts, characteristics and distributions of stars and dust) would reproduce the observations in hand. Computational cost however renders such exercise unfeasible. This project will employ machine learning techniques to develop an emulator capable of mimicking particular functionality of a 3D radiative transfer code in a fraction of the time, and embed it in a statistical framework to enable simultaneous modelling of resolved galaxy information across the electromagnetic spectrum. We will validate the tool on galaxy simulations, and apply it to galaxy observations.
